Summary: Amino acid similarity and structure were used in a model to predict subcellular localizations for prokaryotic and eukaryotic proteins. Tested with Reinhart and Hubbard’s dataset, prediction accuracies reached 1000sing the self-consistency test, 92 0.000000or prokaryotic sequences and 79.2 0.000000or eukaryotic protein sequences with the jackknife test. The results suggest that amino acid structure may be a useful predictor of protein subcellular localization. (Copyright 2004 Online Journal of Bioinformatics)